Cybrpgs/PidginMed-ASR10K

PidginMed-ASR

PidginMed-ASR is a Nigerian Pidgin English medical speech benchmark for automatic speech recognition (ASR).

The dataset contains 10,218 audio samples from 21 speakers, organized into speaker-independent training, validation, and test sets.

Dataset Structure

The dataset contains audio recordings paired with Nigerian Pidgin transcriptions and English source translations.

Column	Description
audio	Path to audio file
transcription	Nigerian Pidgin transcript
english_translation	Original English text
speaker_id	Anonymized speaker ID
duration	Audio duration
split	Train, validation, or test split

Dataset Splits

The dataset uses speaker-independent partitioning, ensuring that no speaker appears in more than one split.

Split	Samples	Speakers	Speaker IDs
Train	6,474	13	3, 5, 7, 8, 9, 11, 12, 13, 15, 16, 18, 19, 20
Validation	1,499	3	2, 14, 21
Test	2,245	5	1, 4, 6, 10, 17
Total	10,218	21	1–21

Note: Speaker IDs follow the naming convention Speaker_X. There is no speaker overlap across training, validation, and test partitions.

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.